The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the bash tool to execute internal scripts and public framework scaffolders.
Evidence includes the invocation of bun new-project/scripts/new-project.ts and bun scripts/extract-skill.ts to manage the project environment and automate pattern extraction.
[EXTERNAL_DOWNLOADS]: The scaffolding process retrieves templates and official packages from external sources via standard package managers and version control.
The engine utilizes git clone for repositories like https://github.com/ariabuilder/aria.git and npx or bun create for various framework and CMS starters (e.g., Next.js, Medusa, Payload CMS).
[INDIRECT_PROMPT_INJECTION]: The skill's 'Client Intake' stage creates a surface where user-provided text data influences the agent's behavior during documentation generation.
Ingestion points: Interactive user responses are stored in Client-Intake/00-Intake-Brief.md and .agents/context/product.md.
Boundary markers: The skill documentation lacks explicit delimiters or 'ignore' instructions for processing user-supplied data in the brief.
Capability inventory: The agent uses bash and write_to_file to transform these inputs into further project documents.
Sanitization: There are no defined filters or sanitization steps for content ingested from the intake brief before it is processed by the agent.
[DATA_EXFILTRATION]: The engine handles sensitive infrastructure information through its 'Technical Intake' process.
Documentation specifies the collection of credentials, repository access tokens, and hosting details in 04-Technical-Intake/ and .env files.
This risk is mitigated by the 'Vibeguard' protocol mentioned in SKILL.md, which uses a pre-commit hook (scripts/pre-commit.sh) to detect and block secrets from being committed to version control.
[DYNAMIC_EXECUTION]: The skill includes functionality for 'Skill Extraction' which involves generating new executable skill files at runtime.
The scripts/extract-skill.ts script allows the agent to generate new SKILL.md contracts and platform definitions based on recurring logic patterns discovered during development.

new-project